Meng-Liang Chung

Visual Contrast Enhancement Algorithm Based on Histogram Equalization

Image enhancement techniques primarily improve the contrast of an image to lend it a better appearance. One of the popular enhancement methods is histogram equalization (HE) because of its simplicity and effectiveness. However, it is rarely applied to consumer electronics products because it can cause excessive contrast enhancement and feature loss problems. These problems make the images processed by HE look unnatural and introduce unwanted artifacts in them. In this study, a visual contrast enhancement algorithm (VCEA) based on HE is proposed. VCEA considers the requirements of the human visual perception in order to address the drawbacks of HE. It effectively solves the excessive contrast enhancement problem by adjusting the spaces between two adjacent gray values of the HE histogram. In addition, VCEA reduces the effects of the feature loss problem by using the obtained spaces. Furthermore, VCEA enhances the detailed textures of an image to generate an enhanced image with better visual quality. Experimental results show that images obtained by applying VCEA have higher contrast and are more suited to human visual perception than those processed by HE and other HE-based methods.

Camera-based Heart Rate measurement using continuous wavelet transform

Recent years have seen increased attention being given to measure Heart Rate with remote-PPG. Previous work mainly focused on optimal signal extraction from three channels of a webcam. Several robust signal extraction methods using either optic models or frequency analysis have been discussed before. After applied these algorithms, residual noise still emerges occasionally with short time duration and relative high energy. This paper investigates the cause of residual noise and then proposed an algorithm based on continuous wavelet transform (CWT) to reduce the highly dynamic residual noise. With noise reduction and signal reconstruction, the advantages of both CWT and FFT are integrated to cope with the motion/illuminance induced residual noise. The results obtained by this system show improved SNR ratio and Entropy under both motion and stationary conditions.

An Embedded Non-Contact Body Temperature Measurement System with Automatic Face Tracking and Neural Network Regression

In the last decade, many advances have been made in the field of automatic temperature estimation, including wearable sensor technologies (WST), infrared thermography (IRT), and non-contact infrared thermometer (NCIT). In contrast with the WST and IRT, NCIT is inexpensive without the risk of potential skin irritation. Nevertheless, NCIT is limited in short valid estimation distance (<12 cm), resulting in the non-satisfaction of the surging application requirements nowadays. This paper proposed an algorithm based on Neural Network Regression not only to reduce the error from 0.6° to 0.12°, which is close to the medical instrument level, but as well to lengthen the valid distance to the range between 50 cm and 100 cm. Furthermore, this study developed an embedded automatic body temperature estimation system which could continuously and unconsciously measure the human temperature in real-time. Integrated with face tracking and fuzzy-control of Pan-tilt unit, the system ensures that human face is focused while measuring. With wireless communication techniques, users can review their physiological Information via App and Web, which is beneficial to remote healthcare.

The applications of automatic vision detection for the intersections

The intelligent transportation systems (ITS) aim to provide services to transport and traffic management and supply more information and safer to various users. The visual-based systems are the most popular solutions for ITS due to their highly maintainable, flexible, and intuitive features. This paper uses a background extraction algorithm to extract initial color backgrounds from surveillance video based on an entropy-analysis concept. The moving objects can then be segmented quickly and correctly by a robust object segmentation algorithm. The segmented object can be used to analyze the trajectory and to provide the collision between pedestrian and moving vehicle. A license plate detection algorithm is also provided to detect the license plate in this study.